README.md

metadata

license: apache-2.0
tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: whisper-small-sp
    results: []

whisper-small-sp

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4485
• Wer: 20.6842

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0005
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 25000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.2671	0.13	1000	2.2108	76.2667
1.4465	0.26	2000	1.6057	67.8753
1.0997	0.39	3000	1.1928	54.2433
0.9389	0.52	4000	1.0020	47.8307
0.7881	0.65	5000	0.8933	46.0046
0.7596	0.78	6000	0.7721	38.5595
0.5678	0.91	7000	0.6903	36.2897
0.4412	1.04	8000	0.6476	32.7473
0.4239	1.17	9000	0.5973	30.8142
0.3935	1.3	10000	0.5444	29.0208
0.3307	1.43	11000	0.5024	27.0434
0.2937	1.56	12000	0.4608	24.7318
0.2471	1.69	13000	0.4259	22.8940
0.2357	1.82	14000	0.3936	21.6018
0.2292	1.95	15000	0.3776	20.8004
0.1493	2.08	16000	0.4599	24.0491
0.1708	2.21	17000	0.4370	23.3443
0.1385	2.34	18000	0.4277	22.3171
0.1288	2.47	19000	0.4050	21.0118
0.1627	2.6	20000	0.4507	23.4004
0.1675	2.73	21000	0.4346	22.8261
0.159	2.86	22000	0.4179	22.2949
0.1458	2.99	23000	0.3978	21.0810
0.0487	3.12	24000	0.4456	20.8617
0.0401	3.25	25000	0.4485	20.6842

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2